Cable Fault Diagnosis Using a Machine Learning Approach Based on Compressed Sensing Time-Domain Reflectometry [Instrumentation and Measurement Systems]

We introduce a novel fault diagnosis method that combines machine learning with the compressed acquisition of CHIRP-Orthogonal Multi-tone Time-Domain Reflectometry (CHIRP-OMTDR). The proposed approach integrates directly into the measurement chain without reconstruction, reducing computational complexity, acquisition bandwidth, and memory requirements, which are key constraints in practical instrumentation systems. The performance of two machine learning models is investigated: the K-Nearest Neighbors (KNN) algorithm and a Convolutional Neural Network (CNN). These models are used separately for impedance and localization estimation, and their performances are compared. Moreover, the impact of the Compression Factor (CF) on fault localization and classification is studied. Our findings demonstrate the effectiveness of machine learning models for directly processing compressed signals, enabling accurate detection, localization, and characterization of faults in cables.